AI Use Case: Beyond Ghibli Filters - Practical Applications of GPT-4o's Image Tools

Have you seen a bunch of people sharing photos of themselves in the Studio Ghibli style and you don't know why? Let me help. OpenAI launched an improvement to their GPT-4o model bringing added capabilities and morphing photos is fun but still the least interesting part about it. The ability to create and manipulate images has taken a dramatic step forward.

OpenAI's upgrade to the GPT-4o model enables true multi-modal support of images. In the past, image generation in many of these models wasn't done natively but instead by farming out individual tasks to image specific models to do the image generation. What this means is that as of this release, the model has more of an understanding of the actual elements of the image. In the same way you can ask ChatGPT or Claude to tweak the introduction of a paragraph, you can now ask it now to remove an element of an image, for instance. What this means practically is you can now engage and manipulate the contents of the images in ways that have been difficult weeks ago and impossible months ago.

I will share 3 areas where you can apply these benefits directly immediately.

1. Product photography
2. Charts and presentation visuals
3. Social Media Campaigns

Product Photography

If you are selling products on the internet, this update could drastically impact your workflow. This makes photoshoots unnecessary for a wide swath of people. It's far from perfect, but this technology is only going to improve moving forward.

For my experimentation purposes, I was playing with a product of a cousin of mine. These products are stockings with a particular cut and styling that makes them very well suited for wearing along with Caribbean carnival costumes. I went to her instagram and took a screen shot of one of her reels, to get an image of the product.

I took that image to ChatGPT and started a chat with GPT-4o. I asked it to:

Give me a commercial product photography hero shot of this product:

No light boxes. No camera. No post processing or color correction.

I then asked it to create a lifestyle image to see the product in a real-life setting. I also tried to contextualize "Carnival" in case it didn't have enough of an understanding already.

Here, admittedly, I ran into some trouble and ChatGPT refused to create a depiction of people in revealing attire. I don't know how much you all are familiar with Carnival anywhere but there is no shortage of "revealing attire". Not a problem all products will face, of course.

I couldn’t generate the requested image because it didn’t meet our content guidelines—specifically around realistic depictions of people in revealing attire. If you’d like, I can still create a lifestyle image that shows the product in a carnival setting, with the packaging featured prominently and festive energy in the background, but without focusing on detailed or revealing outfits. Want to go that route or tweak the concept?

Still, I think it was still able to create something very helpful with more of a Carnival feel than a literal in-use shot.

If you are selling physical products on the internet, you have to be playing with this today. The ability to get an early cut of a creative direction if not generate something wholly passable for with significantly less time and money than existing workflows is just too good to pass up.

Charts and Visuals

The new model offers significantly improved support for text rendering compared to previous iterations. I, myself, have struggled repeatedly trying to generate feature images for posts or charts for presentations that include words with very little success. While it still struggles occasionally with fine text details, the output quality is now consistently reliable enough for professional applications.

For the purposes of this demonstration, I asked it for a visual representing the opportunities represented in this post. I first ask for a "chart visual" and it gave me a bar graph. I had to clarify that a "chart" wasn't necessary and it should focus on the visual.

Words have long been a serious stumbling block for generative AI imagery. Some models have been a bit better than others and you still definitely need to double check all the spellings but this moves these visuals firmly into the area of a legitimate option for work.

Social Media Campaigns

Building on the capabilities demonstrated in product photography and visual design, social media campaigns represent a natural extension of this technology's potential. It can do photos. It can do visuals and text. Right now it exports jpgs so you couldn't consider using it for anything high fidelity like prints but on the internet, this is great.

Once again, my cousin in Jamaica is planning an event and would otherwise spend days going back and forth with a graphic designer to put together collateral for a single revision of an asset for use on social media.

Full disclosure: That little copy of "and BROS" had me walking and talking to myself. The logo had already been used widely elsewhere so I attempted to reproduce it and had varying results. For this image here, I did go into Figma and just white out the garble and replace it with text. Still, this was hours as I refined my process rather than days. And you can continue to generate different models and scenes for an extended content campaign.

The Takeaway

This tool is far from perfect, same as most AI tools. However, it can get you to 80-90% perfect in minutes rather than days or weeks. The investment of money and time are vastly diminished. If your work involves creating visual content for digital platforms, incorporating this technology into your workflow represents a significant competitive advantage. If you still need some help getting started, don't hesitate to reach out.